The example represents a severe performance bug in the bitcoin core software and should be fixed by software changes that make an inefficient operation efficient or restrict very large transactions if this proves too difficult. There is no reason to limit blocksize because of this.
This guy gets it. Unbounded memory use, as well as poorly designed propagation methods are issues that should be addressed by code review, not by adding layers of complexity.
When you look at it from this perspective, you see how the whole 'scaling debate' is being jury rigged to deliver a fundamentally altered version of Bitcoin that suits one particular commercial interest.